Direct Perception & Optic Flow

Week 2: Dynamics of Perception & Action

Tehran Davis

Recap: The Computer vs. The Organism

Traditional View Ecological View
Poverty of Stimulus Rich Information
Retinal Image (Snapshot) Optic Flow (Stream)
Passive Sensation Active Exploration
Inference (Guessing) Detection (Resonance)

The Ambient Optic Array

Gibson’s first move was to redefine the “stimulus.”

  • Not: A flat image on the retina.
  • But: A structured array of light converging at a point.

“The environment is not a collage of images. It is a nested set of solid angles.” — Gibson

Motion is Information

What happens when you move?

  • The array changes.
  • But it doesn’t change randomly. It changes lawfully.
  • This change is called Optic Flow.

The Rules of Locomotion

Gibson proposed simple “Control Laws” based on flow:

  1. To Move Forward: Generate global Outflow (Expansion).
  2. To Move Backward: Generate global Inflow (Contraction).
  3. To Stop: Minimize flow.
  4. To Go to a Target: Place the Focus of Expansion (FOE) on the target.

Evidence: The Moving Room

Lee & Lishman (1975)

  • Subject: Stands on a solid floor.
  • Room: Walls move forward/backward.
  • Result: The subject falls over.

Why?

  • Visual Proprioception: Vision told them they were falling forward, so they leaned back to compensate.

The Problem: “Time to Contact”

Imagine you are a gannet diving into the sea at 60 mph. You need to fold your wings exactly 0.5 seconds before impact. If you fold too early: You crash (too fast). If you fold too late: You break your wings.

How does the bird know when to fold?

The Computational Solution (Hard)

\[ Time = \frac{Distance}{Velocity} \]

To solve this, the brain must estimate: 1. Distance (\(d\)): “I am 50 meters away.” (Hard to judge) 2. Velocity (\(v\)): “I am going 20 m/s.” (Hard to judge) 3. Compute: \(50 / 20 = 2.5\) seconds.

Problem: We are terrible at judging absolute distance and speed.

The Ecological Solution: Tau (\(\tau\))

There is an optical variable available on the retina right now.

  • \(A\): The size of the object’s image on the eye.
  • \(\\dot{A}\): The rate of expansion of that image.

\[ \tau = \frac{A}{\\dot{A}} \]

Rule: The inverse rate of expansion specifies time-to-contact directly.

Why Tau Matters

  1. It is Direct: No inference required.
  2. It is Lawful: It works for birds, bugs, and humans.
  3. It controls Action:
    • Braking: Keep \(\\dot{\\tau}\) constant.
    • Catching: Couple hand closure to \(\\tau\).

Active Perception

“We must perceive in order to move, but we must also move in order to perceive.” — Gibson (1979)

  • Old View: Sensation \(\\rightarrow\) Perception \(\\rightarrow\) Action
  • New View: Perception-Action Loop

We generate the information (flow) that we use to control our behavior.

Next Week: Affordances

We have seen how we detect Events (collisions). Next, we ask how we perceive Possibilities.

Readings: 1. Warren (1984): Stair Climbing 2. Gibson Ch. 8: The Theory of Affordances